Copy page

Guides

Agent skill

Drop2Run publishes a skill file at dropto.run/SKILL.md. It teaches a coding agent how to publish with the CLI: signing in, which folder to point at, and the one rule that matters most — that a deploy with no site named creates a new site rather than updating the last one.

An agent without it works most of this out from drop2run --help. What it does not work out is that rule, and the failure is quiet: "update the site" three times leaves three sites and a URL nobody is looking at.

Installing it

The file is in the open SKILL.md format — YAML frontmatter, then markdown — so the same file works in every agent that reads skills. Only the directory differs.

Codex, Cursor, Gemini CLI and Copilot in VS Code all read .agents/skills/, for one project:

Code
mkdir -p .agents/skills/drop2run
curl -fsSL https://dropto.run/SKILL.md -o .agents/skills/drop2run/SKILL.md

Or ~/.agents/skills/drop2run/ for every project on the machine.

Claude Code, for one project:

Code
mkdir -p .claude/skills/drop2run
curl -fsSL https://dropto.run/SKILL.md -o .claude/skills/drop2run/SKILL.md

Or for every project on the machine:

Code
mkdir -p ~/.claude/skills/drop2run
curl -fsSL https://dropto.run/SKILL.md -o ~/.claude/skills/drop2run/SKILL.md

Nothing else has to be configured. Cursor also reads .claude/skills/, so a project that already has the Claude Code copy does not need a second one there.

Agents without skills: it is one file at a stable URL. Anything that can be pointed at a document — a system prompt, a project instruction file, a retrieval index — can use it. Fetch it rather than copying it in, so an agent gets corrections when they are published.

Keeping it current

Re-run the same curl. There is no version negotiation and no install manifest to keep in step; the file carries a version in its frontmatter, and overwriting it is the whole update procedure.

Code
curl -fsSL https://dropto.run/SKILL.md -o ~/.agents/skills/drop2run/SKILL.md
curl -fsSL https://dropto.run/SKILL.md -o ~/.claude/skills/drop2run/SKILL.md

Run whichever line matches where you installed it.

What it does not include

Credentials. The skill tells an agent to check drop2run whoami and to ask before starting a sign-in, because login opens a browser and waits for a human to approve it. Nothing in the file authenticates anything, and it says explicitly not to print a token and not to try to create one.

Permission to delete. It tells an agent that drop2run rm cannot be undone, that the refusal without --yes should be reported rather than worked around, and not to delete a site that was not named. Those are instructions, not enforcement — the enforcement is in the CLI, which refuses without --yes whatever any agent has been told.

The MCP server's tools. One line points at them — an agent that has the server connected but no CLI installed should call login rather than ask for a global install — and that is all. The tools document themselves to the client that loaded them, so repeating their arguments here would be a second copy that drifts. See Deploy from an AI agent for which integration applies to you.